Echocardiogram View Classification Using Edge Filtered Scale-Invariant Motion Features

1. A method comprising: Obtaining, by a processor, a plurality of video images of a subject; aligning, by the processor, the plurality of images; using the aligned images to generate a motion magnitude image; filtering, by the processor, the motion magnitude image using an edge map; detecting features on the motion magnitude image, discarding those features which do not lie in the neighborhood of edges; encoding, by the processor, the remaining features by: generating, x, y image coordinates for a feature point; generating a motion magnitude histogram in a window around the feature point; and generating a histogram of edge intensity values near the feature point; and using the encoded features to classify the video images of the subject into a predetermined classification.

2. The method of claim 1 wherein said classifying comprises using a vocabulary-based Pyramid Matching Kernel based Support Vector Machine.

3. The method of claim 1 wherein the aligning comprises using affine transformation.

4. The method of claim 1 wherein motion magnitude image is generated using Demons algorithm.

5. The method of claim 1 wherein said video images are echocardiograms.

6. A method of classifying at least one echocardiogram video comprising: representing, by a processor, each image from the echocardiogram video by a set of salient features; modifying, by the processor, the image to produce an edge filtered motion magnitude image; filtering, by the processor, the motion magnitude image using an edge map; detecting features on the motion magnitude image, discarding those features which do not lie in the neighborhood of edges; locating the features at scale invariant points in the edge filtered motion magnitude image; and encoding, by the processor, the edge filtered motion magnitude image by: generating, x, y image coordinate of a feature point; generating a motion magnitude histogram in a window around the feature point; and generating a histogram of edge intensity values near the feature point.

7. The method of claim 6 wherein the encoding comprises encoding the edge filtered motion magnitude image by using spatial information about the image.

8. The method of claim 6 wherein the encoding comprises encoding the edge filtered motion magnitude image by using textual information about the image.

9. The method of claim 6 wherein the encoding comprises encoding the edge filtered motion magnitude image by using kinetic information about the image.

10. The method of claim 6 wherein the locating comprises identifying the scale invariant interest points in motion magnitude that are also close to edges in the edge filtered motion magnitude image.

11. The method of claim 6 wherein the representing comprises representing the image by at least one position (x,y).

12. The method of claim 6 wherein the representing comprises representing the image by at least one histogram of local motion magnitude.

13. The method of claim 6 wherein the representing comprises representing the image by at least one histogram of local intensity.

14. The method of claim 6 wherein the representing comprises representing the image by at least one histogram of local texture.

15. The method of claim 6 further comprising classifying the image into one of a set of predetermined classifications.

16. The method of claim 15 wherein said classifying comprises using a vocabulary-based Pyramid Matching Kernel based Support Vector Machine.